Neural signal drift

Also: nonstationarity, recording instability, neural drift, decoder drift

By neuraspeak editorial · Updated 2026-10-07 · 1 min read

Neural signal drift is the gradual or sudden change in recorded brain signals over hours, days, or months, caused by factors such as tiny electrode movements, tissue reactions, and changes in the brain itself, which makes a BCI decoder trained on earlier data become less accurate.

What causes neural signal drift?

With penetrating electrodes, the brain moves slightly relative to the implant, so the neurons nearest each electrode tip can change. Over longer periods, scar tissue can form around electrodes, and electrode materials can degrade; a multi-year analysis of Utah array recordings in monkeys and people showed that signal quality generally declines with time, though at very different rates across people and arrays (Sponheim et al., Journal of Neural Engineering 2021). The user's brain can also change through learning, fatigue, attention, or disease progression. Within a single session, signal statistics can shift as well.

How do researchers correct for drift?

The simplest fix is recalibration: collecting fresh labeled data and retraining the decoder. Speech BCIs reduce this burden in several ways. Willett et al. (Nature 2023) used a separate input layer for each day to absorb across-day changes and rolling feature adaptation for within-day changes. Card et al. (NEJM 2024) combined data from many days to keep calibrating the decoder, and the participant could trigger a quick recalibration of about 7.5 minutes using 20 prompted sentences.

Another line of work stabilizes decoders without labeled data. Degenhart et al. (Nature Biomedical Engineering 2020) showed in monkeys that aligning the low-dimensional patterns of population activity across days could restore cursor control after severe recording changes, without knowing what the user intended.

Why it matters for speech BCI

For a speech neuroprosthesis to be practical at home, it must work every morning without a research team retraining it. Drift is therefore one of the key barriers between laboratory demonstrations and everyday devices, along with surgical risk and hardware longevity. How well a system tolerates drift, and how much calibration time it needs, is as important to users as peak words per minute.

Questions

Does drift mean the implant is failing?+

Not necessarily. Day-to-day drift is normal and is usually handled by software. A steady long-term loss of signals on many electrodes can reflect hardware or tissue changes.

Do ECoG and EEG systems drift too?+

Yes. All neural recordings are nonstationary to some degree, although the specific causes differ by electrode type. Surface and scalp recordings do not depend on staying near individual neurons.

Sources

Facts on this page are drawn from the sources listed and were accurate to the best of our knowledge on the update date shown. Company and product names are trademarks of their owners and are used for identification only. Nothing here is medical, investment or legal advice. Companies and individuals can request a correction via the corrections page.